Profiles

A profile is a named bundle of Compressor defaults. Profiles are pure convenience — the engine remains generic, language- and domain-neutral. Profiles sit on top so callers can start in one line.

from narrato import Compressor
c = Compressor.from_profile("rag-en")

Built-in profiles

Run narratoflow profiles to print the live list. Current bundle:

name	source_lang	schema	notes
rag-en	en	qa	English RAG; stopword stripping on
qa-en	en	qa	English QA; stopwords preserved for grammar
narrative-en	en	narrative	English narrative generation
interview-en	en	interview	English interview / transcript
dialogue-en	en	dialogue	English scripted dialogue
news-en	en	news	English news article (5W1H)
long-en	en	qa	Long English doc — chunked extraction
narrative-no	no	narrative	Norwegian narrative (original v0.1–v0.2 defaults)
rag-no	no	qa	Norwegian RAG; stopword stripping on

Override profile fields

Any keyword to from_profile() overrides the profile's value:

c = Compressor.from_profile(
    "rag-en",
    provider="openai",
    extractor_model="gpt-4o-mini",
    target_model="gpt-4o",
    source_lang="de",   # override the language to German
)

Register your own

from narrato import Profile, register_profile, Compressor

profile = Profile(
    name="legal-en",
    description="English legal documents — chunked extraction with QA schema",
    source_lang="en",
    schema="qa",
    chunked=True,
    chunk_chars=6000,
    overlap_chars=300,
    extra={"cache": True},   # any other Compressor kwargs go in extra
)
register_profile(profile)

c = Compressor.from_profile("legal-en", provider="anthropic")

Pass overwrite=True to register_profile() to replace an existing entry.

Profile or explicit?

• Use a profile when starting a new use case — saves typing and documents intent.
• Use explicit construction when the profile fields would need so many overrides that the profile no longer reads cleanly.
• Define your own profile if your team will run the same configuration repeatedly.